Insert a user's free-form request into an internal FAQ prompt.
Prevent text in the request from changing system instructions or permissions.
Choose two appropriate controls.
Treat user text as untrusted data, separate it from trusted instructions, validate inputs, and keep authorization outside the model.
Detailed explanation
Input policy can reject or isolate unexpected and malicious content.
Untrusted text should not be treated as a trusted command.
Generated output must be checked by trusted server-side controls.
Input has no authority and permissions require separate authorization.
Check official documentation
Compare the following points with the official documentation.
Amazon Bedrock公式のGuardrails・Prompt management・セキュリティ設計を確認する。
Official references
What to confirm
入力とシステム指示の分離、検証、権限分離をプロンプトインジェクション対策として説明できる。
Key points
- Trust boundary
- Input validation
- Permission separation
Notes
- Environment: AWS公式AIF-C01試験ガイドとAWS公式ドキュメントの確認
- Compare the definition with the AIF-C01 objective and the service documentation before applying it.
- Keep the concept separate from a particular answer choice or implementation step.
Foundation review
Instruction separation
User text should be treated as data and must not gain authority merely because a model reads it.
Independent enforcement
A model's response cannot be the only control for a privileged operation.